Analog Acoustic Feature Extraction for Low-Power Voice Activity Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional voice activity detection (VAD) systems require significant circuit area and power consumption, and are affected by process-voltage-temperature (PVT) variations, necessitating a more efficient solution.
Innovation Solution
An analog VAD system with an acoustic feature extraction (AFE) circuit utilizing band-pass filters and a shared operational amplifier, combined with time-division demultiplexing and multiplexing, and a rectifier, to reduce power consumption and circuit area, while maintaining high performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If digital architecture is used for VAD, then detection accuracy is improved, but power consumption increases and circuit area increases
Solution Approach 1:
The patent replaces the digital signal processing system with an analog signal processing system. The analog VAD circuit processes audio signals directly in analog domain using operational amplifiers, resistors, and capacitors, eliminating the need for analog-to-digital conversion and digital signal processing operations. This substitution significantly reduces power consumption while maintaining detection accuracy through carefully designed analog filtering and rectification circuits that extract voice activity features without requiring substantial computational resources.
Solution Approach 2:
The patent employs a single operational amplifier that serves multiple functions across different frequency channels. The same operational amplifier is time-division multiplexed to perform filtering operations for multiple band-pass filters, and subsequently used for rectification operations. This multi-functional approach reduces the total number of active components required, thereby decreasing circuit area and power consumption while still achieving accurate voice activity detection across multiple frequency bands.
2Measurement precision
If digital architecture is used for VAD, then detection accuracy is improved, but circuit area increases
Solution Approach 1:
The patent replaces the digital signal processing system with an analog signal processing system. The analog VAD circuit processes audio signals directly in analog domain using operational amplifiers, resistors, and capacitors, eliminating the need for analog-to-digital conversion and digital signal processing operations. This substitution significantly reduces power consumption while maintaining detection accuracy through carefully designed analog filtering and rectification circuits that extract voice activity features without requiring substantial computational resources.
Solution Approach 2:
The patent employs a single operational amplifier that serves multiple functions across different frequency channels. The same operational amplifier is time-division multiplexed to perform filtering operations for multiple band-pass filters, and subsequently used for rectification operations. This multi-functional approach reduces the total number of active components required, thereby decreasing circuit area and power consumption while still achieving accurate voice activity detection across multiple frequency bands.
3Measurement precision
If multiple operational amplifiers are used for each channel, then filtering performance is improved, but power consumption increases
Solution Approach 1:
The patent implements periodic action by time-division multiplexing a single operational amplifier across multiple channels and operations. The operational amplifier alternates between serving different band-pass filter channels in different time phases, and then performs rectification operations. This periodic switching allows the system to maintain adequate filtering performance for each channel while using only one operational amplifier, thereby dramatically reducing power consumption compared to having dedicated operational amplifiers for each channel.
Solution Approach 2:
The patent employs dynamic resource allocation where a single operational amplifier dynamically switches between different functions and channels based on timing signals. The operational amplifier's role changes over time - sometimes acting as a filter for different frequency bands, other times acting as a rectifier. This dynamic usage pattern allows the system to achieve multi-channel filtering performance with a single operational amplifier, reducing power consumption while maintaining the necessary filtering quality for voice activity detection.
Data Source
AI summary
An acoustic feature extraction (AFE) circuit includes a plurality of band-pass filters (BPFs) adaptable to a plurality of channels with different band-pass frequency ranges respectively for switchably receiving an amplified signal, thereby generating corresponding filtered signals, the plurality of BPFs including an operational amplifier that is shared among the plurality of channels; and a rectifier switchably coupled to receive the filtered signals, thereby generating a rectified signal. The amplified signal is time-division demultiplexed onto the BPFs in different phases, and the filtered signals are time-division multiplexed onto the rectifier in different phases.